Instead of distributing the driver's bookkeeping over 5 (i.e.
TPU_CHANNEL_MAX + 1) separately allocated memory chunks, put all together
in struct tpu_device. This reduces the number of memory allocations and
so fragmentation and maybe even the number of cache misses. Also
&tpu->tpd[pwm->hwpwm] is cheaper to evaluate than pwm_get_chip_data(pwm)
as the former is just an addition in machine code while the latter involves
a function call.

Link: https://lore.kernel.org/r/20230705080650.2353391-6-u.kleine-koenig@pengutronix.deSigned-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Thierry Reding <thierry.reding@gmail.com>